Many current automatic speech recognition (ASR) systems require a user to explicitly train acoustic models by reading predetermined sentences to adapt a speaker-independent (SI) acoustic model based on speech characteristics of the user to improve speech recognition performance.